Spatial Hierarchies and Latent Preferences: Rethinking Submarket Boundaries in Urban Housing

This paper presents a proof-of-concept comparison of two approaches for delineating intra-urban housing submarkets, using Atlanta, Georgia as an illustrative case. The first approach applies the hierarchical model of Goodman and Thibodeau, which defines submarkets as spatially contiguous areas nested within high school districts. The second employs a fully endogenized finite mixture model (FMM) that partitions home sales into latent groups of households with similar preferences, independent of geographic location. Using approximately 5,000 home sales from 2015-2016, a relatively stable period in the local housing market, we examine how each approach organizes house price variation and residential sorting, rather than attempting to identify a single optimal submarket structure. Both models generate efficient hedonic price estimates, but they differ fundamentally in the economic insights they provide. The hierarchical model offers a stable spatial framework for estimating the in situ capitalization of neighborhood amenities and public goods such as school quality. By contrast, the FMM reveals latent preference structures and highlights household sorting across potentially non-adjacent neighborhoods. Taken together, the results illustrate the complementary roles of spatially contiguous and preference-based submarket definitions in housing market analysis. Rather than advocating one approach over the other, the paper clarifies how different submarket strategies serve distinct analytical and policy objectives.